Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
The writer is a managing director at Panmure Liberum
US stock market valuations are once again approaching the highs of the 2000 dotcom bubble. This already points to the increased risk of significant losses in the future, but what makes things worse is that these extreme valuations are being achieved at a time when company earnings are also well above normal levels. If valuations normalise and earnings decline at the same time, the losses for investors could compound rapidly.
Professor Robert Shiller from Yale University popularised the cyclically adjusted price-to-earnings ratio (CAPE) as a measure for the long-term prospects of the S&P 500. This ratio is back in the spotlight after it surpassed 40 times in May, the first and only time this has happened since the peak of the tech bubble of the late 1990s.
This invited the obvious comparisons between the current boom driven by the revolutionary potential of AI and the boom of the 1990s driven by the revolutionary potential of the internet.
I happen to believe that we are once again in a period of over-investment in technology. But one counter-argument that is often made against a comparison of the AI boom with the dotcom boom is that the hyperscalers and semiconductor companies that fuel the current boom are highly profitable, unlike the internet start-ups of 30-odd years ago.
But I doubt that this is such a good argument to begin with. These companies are not just profitable — they currently earn supernormal profits. If I fit the earnings per share of the S&P 500 as reported by Shiller with an exponential growth trend, then current earnings are about 59 per cent above trend. This is up from 14 per cent above trend at the start of 2023 when the AI boom began. Indeed, most of the earnings acceleration came in the past 12 months.
We can compare current and past valuations and earnings by normalising them into a z-score, a common measure in science and engineering that allows us to compare different variables measured at different scales. In our case, the z-score compares the CAPE and earnings per share of the S&P 500 to the long-term average and divides the difference by historic volatility. A z-score of +1 implies that the current value is one standard deviation above normal and values above +2 are typically considered dangerously high or even a bubble.
The CAPE of the S&P 500 currently has a z-score of 2.9, which is well in bubble territory. It isn’t as extreme as the 3.3 reading in December 1999, but not far off either. And it is significantly higher than the z-score of 1.8 at the height of the 1929 stock market bubble.
But the real difference between today’s market boom and the peak of the dotcom bubble lies in the z-score of S&P 500 earnings. In early 2000, US earnings per share showed a z-score of 0.9. Elevated, but nothing too concerning. Plus, earnings were elevated in the old economy at large, not in the booming tech stocks. Today, the z-score of S&P 500 earnings per share is 1.8, almost in bubble territory. And these earnings are highly concentrated in the companies associated with the AI boom.
Some content could not load. Check your internet connection or browser settings.
The upshot is that we are in the middle of a valuation bubble while approaching an earnings bubble. Current supernormal earnings may be sustained for years to come. The Magnificent Seven and other large tech stocks have managed to keep profit margins and earnings at elevated levels for more than a decade because they are capital-light businesses with product offerings that customers cannot easily replace.
But if earnings were to normalise and fall back to their long-term trend, which appears increasingly likely as tech companies rapidly increase their capital expenditures, the current CAPE of the S&P 500 would not be 40 times but about 64 times. And that would be a z-score of 4.6.
Valuations don’t follow a normal distribution but for the sake of illustration, we can for a moment assume they do. In this case a z-score of 4.6 corresponds to an event that happens in 0.00019 per cent of months or only once in 43,432 years. We truly live in exceptional times in the US stock market.
Of course, this is not to say that it will crash immediately. The experience of the late 1990s tells us that tech booms can last longer than many people expect. But the analysis tells us that for long-term investors, risks outweigh the opportunities. While I do not advocate selling US tech stocks or US stocks in general, I believe investors need to stay vigilant and be ready to reduce positions when the tide turns.
